]>
Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
2817b000 SW |
2 | |
3 | #include <linux/spinlock.h> | |
4 | #include <linux/fs_struct.h> | |
5 | #include <linux/namei.h> | |
5a0e3ad6 | 6 | #include <linux/slab.h> |
2817b000 SW |
7 | #include <linux/sched.h> |
8 | ||
9 | #include "super.h" | |
3d14c5d2 | 10 | #include "mds_client.h" |
2817b000 SW |
11 | |
12 | /* | |
13 | * Directory operations: readdir, lookup, create, link, unlink, | |
14 | * rename, etc. | |
15 | */ | |
16 | ||
17 | /* | |
18 | * Ceph MDS operations are specified in terms of a base ino and | |
19 | * relative path. Thus, the client can specify an operation on a | |
20 | * specific inode (e.g., a getattr due to fstat(2)), or as a path | |
21 | * relative to, say, the root directory. | |
22 | * | |
23 | * Normally, we limit ourselves to strict inode ops (no path component) | |
24 | * or dentry operations (a single path component relative to an ino). The | |
25 | * exception to this is open_root_dentry(), which will open the mount | |
26 | * point by name. | |
27 | */ | |
28 | ||
29 | const struct inode_operations ceph_dir_iops; | |
30 | const struct file_operations ceph_dir_fops; | |
52dfb8ac | 31 | const struct dentry_operations ceph_dentry_ops; |
2817b000 SW |
32 | |
33 | /* | |
34 | * Initialize ceph dentry state. | |
35 | */ | |
36 | int ceph_init_dentry(struct dentry *dentry) | |
37 | { | |
38 | struct ceph_dentry_info *di; | |
39 | ||
40 | if (dentry->d_fsdata) | |
41 | return 0; | |
42 | ||
36e21687 | 43 | di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO); |
2817b000 SW |
44 | if (!di) |
45 | return -ENOMEM; /* oh well */ | |
46 | ||
47 | spin_lock(&dentry->d_lock); | |
8c6efb58 SW |
48 | if (dentry->d_fsdata) { |
49 | /* lost a race */ | |
50 | kmem_cache_free(ceph_dentry_cachep, di); | |
2817b000 | 51 | goto out_unlock; |
8c6efb58 | 52 | } |
48d0cbd1 | 53 | |
8842b3be | 54 | if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) |
48d0cbd1 SW |
55 | d_set_d_op(dentry, &ceph_dentry_ops); |
56 | else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR) | |
57 | d_set_d_op(dentry, &ceph_snapdir_dentry_ops); | |
58 | else | |
59 | d_set_d_op(dentry, &ceph_snap_dentry_ops); | |
60 | ||
2817b000 SW |
61 | di->dentry = dentry; |
62 | di->lease_session = NULL; | |
2817b000 | 63 | dentry->d_time = jiffies; |
48d0cbd1 SW |
64 | /* avoid reordering d_fsdata setup so that the check above is safe */ |
65 | smp_mb(); | |
66 | dentry->d_fsdata = di; | |
2817b000 SW |
67 | ceph_dentry_lru_add(dentry); |
68 | out_unlock: | |
69 | spin_unlock(&dentry->d_lock); | |
70 | return 0; | |
71 | } | |
72 | ||
5f21c96d SW |
73 | struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry) |
74 | { | |
75 | struct inode *inode = NULL; | |
76 | ||
77 | if (!dentry) | |
78 | return NULL; | |
79 | ||
80 | spin_lock(&dentry->d_lock); | |
8842b3be | 81 | if (!IS_ROOT(dentry)) { |
5f21c96d SW |
82 | inode = dentry->d_parent->d_inode; |
83 | ihold(inode); | |
84 | } | |
85 | spin_unlock(&dentry->d_lock); | |
86 | return inode; | |
87 | } | |
2817b000 SW |
88 | |
89 | ||
90 | /* | |
91 | * for readdir, we encode the directory frag and offset within that | |
92 | * frag into f_pos. | |
93 | */ | |
94 | static unsigned fpos_frag(loff_t p) | |
95 | { | |
96 | return p >> 32; | |
97 | } | |
98 | static unsigned fpos_off(loff_t p) | |
99 | { | |
100 | return p & 0xffffffff; | |
101 | } | |
102 | ||
4d5f5df6 YZ |
103 | static int fpos_cmp(loff_t l, loff_t r) |
104 | { | |
105 | int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r)); | |
106 | if (v) | |
107 | return v; | |
108 | return (int)(fpos_off(l) - fpos_off(r)); | |
109 | } | |
110 | ||
2817b000 SW |
111 | /* |
112 | * When possible, we try to satisfy a readdir by peeking at the | |
113 | * dcache. We make this work by carefully ordering dentries on | |
114 | * d_u.d_child when we initially get results back from the MDS, and | |
115 | * falling back to a "normal" sync readdir if any dentries in the dir | |
116 | * are dropped. | |
117 | * | |
2f276c51 | 118 | * Complete dir indicates that we have all dentries in the dir. It is |
2817b000 SW |
119 | * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by |
120 | * the MDS if/when the directory is modified). | |
121 | */ | |
a30be7cb YZ |
122 | static int __dcache_readdir(struct file *file, struct dir_context *ctx, |
123 | u32 shared_gen) | |
2817b000 | 124 | { |
77acfa29 AV |
125 | struct ceph_file_info *fi = file->private_data; |
126 | struct dentry *parent = file->f_dentry; | |
2817b000 SW |
127 | struct inode *dir = parent->d_inode; |
128 | struct list_head *p; | |
129 | struct dentry *dentry, *last; | |
130 | struct ceph_dentry_info *di; | |
131 | int err = 0; | |
132 | ||
133 | /* claim ref on last dentry we returned */ | |
134 | last = fi->dentry; | |
135 | fi->dentry = NULL; | |
136 | ||
a30be7cb YZ |
137 | dout("__dcache_readdir %p v%u at %llu (last %p)\n", |
138 | dir, shared_gen, ctx->pos, last); | |
2817b000 | 139 | |
2fd6b7f5 | 140 | spin_lock(&parent->d_lock); |
2817b000 SW |
141 | |
142 | /* start at beginning? */ | |
77acfa29 | 143 | if (ctx->pos == 2 || last == NULL || |
6da5246d | 144 | fpos_cmp(ctx->pos, ceph_dentry(last)->offset) < 0) { |
2817b000 SW |
145 | if (list_empty(&parent->d_subdirs)) |
146 | goto out_unlock; | |
147 | p = parent->d_subdirs.prev; | |
148 | dout(" initial p %p/%p\n", p->prev, p->next); | |
149 | } else { | |
150 | p = last->d_u.d_child.prev; | |
151 | } | |
152 | ||
153 | more: | |
154 | dentry = list_entry(p, struct dentry, d_u.d_child); | |
155 | di = ceph_dentry(dentry); | |
156 | while (1) { | |
1cd3935b SW |
157 | dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next, |
158 | d_unhashed(dentry) ? "!hashed" : "hashed", | |
2817b000 SW |
159 | parent->d_subdirs.prev, parent->d_subdirs.next); |
160 | if (p == &parent->d_subdirs) { | |
9cfa1098 | 161 | fi->flags |= CEPH_F_ATEND; |
2817b000 SW |
162 | goto out_unlock; |
163 | } | |
2fd6b7f5 | 164 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
a30be7cb YZ |
165 | if (di->lease_shared_gen == shared_gen && |
166 | !d_unhashed(dentry) && dentry->d_inode && | |
09b8a7d2 | 167 | ceph_snap(dentry->d_inode) != CEPH_SNAPDIR && |
1d1de916 | 168 | ceph_ino(dentry->d_inode) != CEPH_INO_CEPH && |
4d5f5df6 | 169 | fpos_cmp(ctx->pos, di->offset) <= 0) |
2817b000 SW |
170 | break; |
171 | dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry, | |
172 | dentry->d_name.len, dentry->d_name.name, di->offset, | |
77acfa29 | 173 | ctx->pos, d_unhashed(dentry) ? " unhashed" : "", |
2817b000 | 174 | !dentry->d_inode ? " null" : ""); |
da502956 | 175 | spin_unlock(&dentry->d_lock); |
2817b000 SW |
176 | p = p->prev; |
177 | dentry = list_entry(p, struct dentry, d_u.d_child); | |
178 | di = ceph_dentry(dentry); | |
179 | } | |
180 | ||
da502956 | 181 | dget_dlock(dentry); |
b7ab39f6 | 182 | spin_unlock(&dentry->d_lock); |
2fd6b7f5 | 183 | spin_unlock(&parent->d_lock); |
2817b000 | 184 | |
0081bd83 YZ |
185 | /* make sure a dentry wasn't dropped while we didn't have parent lock */ |
186 | if (!ceph_dir_is_complete(dir)) { | |
187 | dout(" lost dir complete on %p; falling back to mds\n", dir); | |
188 | dput(dentry); | |
189 | err = -EAGAIN; | |
190 | goto out; | |
191 | } | |
192 | ||
77acfa29 | 193 | dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, ctx->pos, |
2817b000 | 194 | dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode); |
77acfa29 AV |
195 | if (!dir_emit(ctx, dentry->d_name.name, |
196 | dentry->d_name.len, | |
ad1fee96 | 197 | ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino), |
77acfa29 AV |
198 | dentry->d_inode->i_mode >> 12)) { |
199 | if (last) { | |
2817b000 SW |
200 | /* remember our position */ |
201 | fi->dentry = last; | |
f0494206 | 202 | fi->next_offset = fpos_off(di->offset); |
2817b000 | 203 | } |
77acfa29 AV |
204 | dput(dentry); |
205 | return 0; | |
2817b000 | 206 | } |
f5b06628 | 207 | |
0081bd83 YZ |
208 | ctx->pos = di->offset + 1; |
209 | ||
77acfa29 AV |
210 | if (last) |
211 | dput(last); | |
212 | last = dentry; | |
2817b000 | 213 | |
2fd6b7f5 | 214 | spin_lock(&parent->d_lock); |
efa4c120 SW |
215 | p = p->prev; /* advance to next dentry */ |
216 | goto more; | |
2817b000 SW |
217 | |
218 | out_unlock: | |
2fd6b7f5 | 219 | spin_unlock(&parent->d_lock); |
efa4c120 SW |
220 | out: |
221 | if (last) | |
2817b000 | 222 | dput(last); |
2817b000 SW |
223 | return err; |
224 | } | |
225 | ||
226 | /* | |
227 | * make note of the last dentry we read, so we can | |
228 | * continue at the same lexicographical point, | |
229 | * regardless of what dir changes take place on the | |
230 | * server. | |
231 | */ | |
232 | static int note_last_dentry(struct ceph_file_info *fi, const char *name, | |
233 | int len) | |
234 | { | |
235 | kfree(fi->last_name); | |
236 | fi->last_name = kmalloc(len+1, GFP_NOFS); | |
237 | if (!fi->last_name) | |
238 | return -ENOMEM; | |
239 | memcpy(fi->last_name, name, len); | |
240 | fi->last_name[len] = 0; | |
241 | dout("note_last_dentry '%s'\n", fi->last_name); | |
242 | return 0; | |
243 | } | |
244 | ||
77acfa29 | 245 | static int ceph_readdir(struct file *file, struct dir_context *ctx) |
2817b000 | 246 | { |
77acfa29 AV |
247 | struct ceph_file_info *fi = file->private_data; |
248 | struct inode *inode = file_inode(file); | |
2817b000 | 249 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
250 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
251 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
77acfa29 AV |
252 | unsigned frag = fpos_frag(ctx->pos); |
253 | int off = fpos_off(ctx->pos); | |
2817b000 SW |
254 | int err; |
255 | u32 ftype; | |
256 | struct ceph_mds_reply_info_parsed *rinfo; | |
2817b000 | 257 | |
77acfa29 | 258 | dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off); |
9cfa1098 | 259 | if (fi->flags & CEPH_F_ATEND) |
2817b000 SW |
260 | return 0; |
261 | ||
262 | /* always start with . and .. */ | |
77acfa29 | 263 | if (ctx->pos == 0) { |
2817b000 SW |
264 | /* note dir version at start of readdir so we can tell |
265 | * if any dentries get dropped */ | |
2f276c51 | 266 | fi->dir_release_count = atomic_read(&ci->i_release_count); |
2817b000 SW |
267 | |
268 | dout("readdir off 0 -> '.'\n"); | |
77acfa29 | 269 | if (!dir_emit(ctx, ".", 1, |
ad1fee96 | 270 | ceph_translate_ino(inode->i_sb, inode->i_ino), |
77acfa29 | 271 | inode->i_mode >> 12)) |
2817b000 | 272 | return 0; |
77acfa29 | 273 | ctx->pos = 1; |
2817b000 SW |
274 | off = 1; |
275 | } | |
77acfa29 AV |
276 | if (ctx->pos == 1) { |
277 | ino_t ino = parent_ino(file->f_dentry); | |
2817b000 | 278 | dout("readdir off 1 -> '..'\n"); |
77acfa29 | 279 | if (!dir_emit(ctx, "..", 2, |
ad1fee96 | 280 | ceph_translate_ino(inode->i_sb, ino), |
77acfa29 | 281 | inode->i_mode >> 12)) |
2817b000 | 282 | return 0; |
77acfa29 | 283 | ctx->pos = 2; |
2817b000 SW |
284 | off = 2; |
285 | } | |
286 | ||
287 | /* can we use the dcache? */ | |
be655596 | 288 | spin_lock(&ci->i_ceph_lock); |
77acfa29 | 289 | if ((ctx->pos == 2 || fi->dentry) && |
3d14c5d2 | 290 | !ceph_test_mount_opt(fsc, NOASYNCREADDIR) && |
a0dff78d | 291 | ceph_snap(inode) != CEPH_SNAPDIR && |
2f276c51 | 292 | __ceph_dir_is_complete(ci) && |
2817b000 | 293 | __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) { |
a30be7cb | 294 | u32 shared_gen = ci->i_shared_gen; |
be655596 | 295 | spin_unlock(&ci->i_ceph_lock); |
a30be7cb | 296 | err = __dcache_readdir(file, ctx, shared_gen); |
efa4c120 | 297 | if (err != -EAGAIN) |
2817b000 | 298 | return err; |
0081bd83 YZ |
299 | frag = fpos_frag(ctx->pos); |
300 | off = fpos_off(ctx->pos); | |
efa4c120 | 301 | } else { |
be655596 | 302 | spin_unlock(&ci->i_ceph_lock); |
2817b000 | 303 | } |
2817b000 SW |
304 | if (fi->dentry) { |
305 | err = note_last_dentry(fi, fi->dentry->d_name.name, | |
306 | fi->dentry->d_name.len); | |
307 | if (err) | |
308 | return err; | |
309 | dput(fi->dentry); | |
310 | fi->dentry = NULL; | |
311 | } | |
312 | ||
313 | /* proceed with a normal readdir */ | |
314 | ||
315 | more: | |
316 | /* do we have the correct frag content buffered? */ | |
317 | if (fi->frag != frag || fi->last_readdir == NULL) { | |
318 | struct ceph_mds_request *req; | |
319 | int op = ceph_snap(inode) == CEPH_SNAPDIR ? | |
320 | CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR; | |
321 | ||
322 | /* discard old result, if any */ | |
393f6620 | 323 | if (fi->last_readdir) { |
2817b000 | 324 | ceph_mdsc_put_request(fi->last_readdir); |
393f6620 SW |
325 | fi->last_readdir = NULL; |
326 | } | |
2817b000 | 327 | |
2817b000 SW |
328 | dout("readdir fetching %llx.%llx frag %x offset '%s'\n", |
329 | ceph_vinop(inode), frag, fi->last_name); | |
330 | req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); | |
331 | if (IS_ERR(req)) | |
332 | return PTR_ERR(req); | |
54008399 YZ |
333 | err = ceph_alloc_readdir_reply_buffer(req, inode); |
334 | if (err) { | |
335 | ceph_mdsc_put_request(req); | |
336 | return err; | |
337 | } | |
70b666c3 SW |
338 | req->r_inode = inode; |
339 | ihold(inode); | |
77acfa29 | 340 | req->r_dentry = dget(file->f_dentry); |
2817b000 SW |
341 | /* hints to request -> mds selection code */ |
342 | req->r_direct_mode = USE_AUTH_MDS; | |
343 | req->r_direct_hash = ceph_frag_value(frag); | |
344 | req->r_direct_is_hash = true; | |
345 | req->r_path2 = kstrdup(fi->last_name, GFP_NOFS); | |
346 | req->r_readdir_offset = fi->next_offset; | |
347 | req->r_args.readdir.frag = cpu_to_le32(frag); | |
2817b000 SW |
348 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
349 | if (err < 0) { | |
350 | ceph_mdsc_put_request(req); | |
351 | return err; | |
352 | } | |
353 | dout("readdir got and parsed readdir result=%d" | |
354 | " on frag %x, end=%d, complete=%d\n", err, frag, | |
355 | (int)req->r_reply_info.dir_end, | |
356 | (int)req->r_reply_info.dir_complete); | |
357 | ||
358 | if (!req->r_did_prepopulate) { | |
359 | dout("readdir !did_prepopulate"); | |
2f276c51 YZ |
360 | /* preclude from marking dir complete */ |
361 | fi->dir_release_count--; | |
2817b000 SW |
362 | } |
363 | ||
364 | /* note next offset and last dentry name */ | |
81c6aea5 YZ |
365 | rinfo = &req->r_reply_info; |
366 | if (le32_to_cpu(rinfo->dir_dir->frag) != frag) { | |
367 | frag = le32_to_cpu(rinfo->dir_dir->frag); | |
368 | if (ceph_frag_is_leftmost(frag)) | |
369 | fi->next_offset = 2; | |
370 | else | |
371 | fi->next_offset = 0; | |
372 | off = fi->next_offset; | |
373 | } | |
f0494206 | 374 | fi->frag = frag; |
2817b000 SW |
375 | fi->offset = fi->next_offset; |
376 | fi->last_readdir = req; | |
377 | ||
378 | if (req->r_reply_info.dir_end) { | |
379 | kfree(fi->last_name); | |
380 | fi->last_name = NULL; | |
7b88dadc SW |
381 | if (ceph_frag_is_rightmost(frag)) |
382 | fi->next_offset = 2; | |
383 | else | |
384 | fi->next_offset = 0; | |
2817b000 | 385 | } else { |
2817b000 SW |
386 | err = note_last_dentry(fi, |
387 | rinfo->dir_dname[rinfo->dir_nr-1], | |
388 | rinfo->dir_dname_len[rinfo->dir_nr-1]); | |
389 | if (err) | |
390 | return err; | |
391 | fi->next_offset += rinfo->dir_nr; | |
392 | } | |
393 | } | |
394 | ||
395 | rinfo = &fi->last_readdir->r_reply_info; | |
396 | dout("readdir frag %x num %d off %d chunkoff %d\n", frag, | |
397 | rinfo->dir_nr, off, fi->offset); | |
77acfa29 AV |
398 | |
399 | ctx->pos = ceph_make_fpos(frag, off); | |
da39822c | 400 | while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) { |
2817b000 SW |
401 | struct ceph_mds_reply_inode *in = |
402 | rinfo->dir_in[off - fi->offset].in; | |
3105c19c SW |
403 | struct ceph_vino vino; |
404 | ino_t ino; | |
405 | ||
2817b000 | 406 | dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n", |
77acfa29 | 407 | off, off - fi->offset, rinfo->dir_nr, ctx->pos, |
2817b000 SW |
408 | rinfo->dir_dname_len[off - fi->offset], |
409 | rinfo->dir_dname[off - fi->offset], in); | |
410 | BUG_ON(!in); | |
411 | ftype = le32_to_cpu(in->mode) >> 12; | |
3105c19c SW |
412 | vino.ino = le64_to_cpu(in->ino); |
413 | vino.snap = le64_to_cpu(in->snapid); | |
414 | ino = ceph_vino_to_ino(vino); | |
77acfa29 | 415 | if (!dir_emit(ctx, |
2817b000 SW |
416 | rinfo->dir_dname[off - fi->offset], |
417 | rinfo->dir_dname_len[off - fi->offset], | |
77acfa29 | 418 | ceph_translate_ino(inode->i_sb, ino), ftype)) { |
2817b000 SW |
419 | dout("filldir stopping us...\n"); |
420 | return 0; | |
421 | } | |
422 | off++; | |
77acfa29 | 423 | ctx->pos++; |
2817b000 SW |
424 | } |
425 | ||
426 | if (fi->last_name) { | |
427 | ceph_mdsc_put_request(fi->last_readdir); | |
428 | fi->last_readdir = NULL; | |
429 | goto more; | |
430 | } | |
431 | ||
432 | /* more frags? */ | |
433 | if (!ceph_frag_is_rightmost(frag)) { | |
434 | frag = ceph_frag_next(frag); | |
435 | off = 0; | |
77acfa29 | 436 | ctx->pos = ceph_make_fpos(frag, off); |
2817b000 SW |
437 | dout("readdir next frag is %x\n", frag); |
438 | goto more; | |
439 | } | |
9cfa1098 | 440 | fi->flags |= CEPH_F_ATEND; |
2817b000 SW |
441 | |
442 | /* | |
443 | * if dir_release_count still matches the dir, no dentries | |
444 | * were released during the whole readdir, and we should have | |
445 | * the complete dir contents in our cache. | |
446 | */ | |
be655596 | 447 | spin_lock(&ci->i_ceph_lock); |
2f276c51 | 448 | if (atomic_read(&ci->i_release_count) == fi->dir_release_count) { |
a8673d61 | 449 | dout(" marking %p complete\n", inode); |
2f276c51 | 450 | __ceph_dir_set_complete(ci, fi->dir_release_count); |
2817b000 | 451 | } |
be655596 | 452 | spin_unlock(&ci->i_ceph_lock); |
2817b000 | 453 | |
77acfa29 | 454 | dout("readdir %p file %p done.\n", inode, file); |
2817b000 SW |
455 | return 0; |
456 | } | |
457 | ||
dcd3cc05 | 458 | static void reset_readdir(struct ceph_file_info *fi, unsigned frag) |
2817b000 SW |
459 | { |
460 | if (fi->last_readdir) { | |
461 | ceph_mdsc_put_request(fi->last_readdir); | |
462 | fi->last_readdir = NULL; | |
463 | } | |
464 | kfree(fi->last_name); | |
a1629c3b | 465 | fi->last_name = NULL; |
dcd3cc05 YZ |
466 | if (ceph_frag_is_leftmost(frag)) |
467 | fi->next_offset = 2; /* compensate for . and .. */ | |
468 | else | |
469 | fi->next_offset = 0; | |
2817b000 SW |
470 | if (fi->dentry) { |
471 | dput(fi->dentry); | |
472 | fi->dentry = NULL; | |
473 | } | |
9cfa1098 | 474 | fi->flags &= ~CEPH_F_ATEND; |
2817b000 SW |
475 | } |
476 | ||
965c8e59 | 477 | static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence) |
2817b000 SW |
478 | { |
479 | struct ceph_file_info *fi = file->private_data; | |
480 | struct inode *inode = file->f_mapping->host; | |
f0494206 | 481 | loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset); |
2817b000 SW |
482 | loff_t retval; |
483 | ||
484 | mutex_lock(&inode->i_mutex); | |
06222e49 | 485 | retval = -EINVAL; |
965c8e59 | 486 | switch (whence) { |
2817b000 SW |
487 | case SEEK_END: |
488 | offset += inode->i_size + 2; /* FIXME */ | |
489 | break; | |
490 | case SEEK_CUR: | |
491 | offset += file->f_pos; | |
06222e49 JB |
492 | case SEEK_SET: |
493 | break; | |
494 | default: | |
495 | goto out; | |
2817b000 | 496 | } |
06222e49 | 497 | |
f0494206 | 498 | if (offset >= 0) { |
2817b000 SW |
499 | if (offset != file->f_pos) { |
500 | file->f_pos = offset; | |
501 | file->f_version = 0; | |
9cfa1098 | 502 | fi->flags &= ~CEPH_F_ATEND; |
2817b000 SW |
503 | } |
504 | retval = offset; | |
505 | ||
506 | /* | |
507 | * discard buffered readdir content on seekdir(0), or | |
508 | * seek to new frag, or seek prior to current chunk. | |
509 | */ | |
510 | if (offset == 0 || | |
f0494206 | 511 | fpos_frag(offset) != fi->frag || |
2817b000 SW |
512 | fpos_off(offset) < fi->offset) { |
513 | dout("dir_llseek dropping %p content\n", file); | |
dcd3cc05 | 514 | reset_readdir(fi, fpos_frag(offset)); |
2817b000 SW |
515 | } |
516 | ||
517 | /* bump dir_release_count if we did a forward seek */ | |
f0494206 | 518 | if (fpos_cmp(offset, old_offset) > 0) |
2817b000 SW |
519 | fi->dir_release_count--; |
520 | } | |
06222e49 | 521 | out: |
2817b000 SW |
522 | mutex_unlock(&inode->i_mutex); |
523 | return retval; | |
524 | } | |
525 | ||
526 | /* | |
468640e3 | 527 | * Handle lookups for the hidden .snap directory. |
2817b000 | 528 | */ |
468640e3 SW |
529 | int ceph_handle_snapdir(struct ceph_mds_request *req, |
530 | struct dentry *dentry, int err) | |
2817b000 | 531 | { |
3d14c5d2 | 532 | struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb); |
d79698da | 533 | struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */ |
2817b000 SW |
534 | |
535 | /* .snap dir? */ | |
536 | if (err == -ENOENT && | |
455cec0a | 537 | ceph_snap(parent) == CEPH_NOSNAP && |
6b805185 | 538 | strcmp(dentry->d_name.name, |
3d14c5d2 | 539 | fsc->mount_options->snapdir_name) == 0) { |
2817b000 SW |
540 | struct inode *inode = ceph_get_snapdir(parent); |
541 | dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n", | |
542 | dentry, dentry->d_name.len, dentry->d_name.name, inode); | |
9358c6d4 | 543 | BUG_ON(!d_unhashed(dentry)); |
2817b000 SW |
544 | d_add(dentry, inode); |
545 | err = 0; | |
546 | } | |
468640e3 SW |
547 | return err; |
548 | } | |
2817b000 | 549 | |
468640e3 SW |
550 | /* |
551 | * Figure out final result of a lookup/open request. | |
552 | * | |
553 | * Mainly, make sure we return the final req->r_dentry (if it already | |
554 | * existed) in place of the original VFS-provided dentry when they | |
555 | * differ. | |
556 | * | |
557 | * Gracefully handle the case where the MDS replies with -ENOENT and | |
558 | * no trace (which it may do, at its discretion, e.g., if it doesn't | |
559 | * care to issue a lease on the negative dentry). | |
560 | */ | |
561 | struct dentry *ceph_finish_lookup(struct ceph_mds_request *req, | |
562 | struct dentry *dentry, int err) | |
563 | { | |
2817b000 SW |
564 | if (err == -ENOENT) { |
565 | /* no trace? */ | |
566 | err = 0; | |
567 | if (!req->r_reply_info.head->is_dentry) { | |
568 | dout("ENOENT and no trace, dentry %p inode %p\n", | |
569 | dentry, dentry->d_inode); | |
570 | if (dentry->d_inode) { | |
571 | d_drop(dentry); | |
572 | err = -ENOENT; | |
573 | } else { | |
574 | d_add(dentry, NULL); | |
575 | } | |
576 | } | |
577 | } | |
578 | if (err) | |
579 | dentry = ERR_PTR(err); | |
580 | else if (dentry != req->r_dentry) | |
581 | dentry = dget(req->r_dentry); /* we got spliced */ | |
582 | else | |
583 | dentry = NULL; | |
584 | return dentry; | |
585 | } | |
586 | ||
1d1de916 SW |
587 | static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry) |
588 | { | |
589 | return ceph_ino(inode) == CEPH_INO_ROOT && | |
590 | strncmp(dentry->d_name.name, ".ceph", 5) == 0; | |
591 | } | |
592 | ||
2817b000 SW |
593 | /* |
594 | * Look up a single dir entry. If there is a lookup intent, inform | |
595 | * the MDS so that it gets our 'caps wanted' value in a single op. | |
596 | */ | |
597 | static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry, | |
00cd8dd3 | 598 | unsigned int flags) |
2817b000 | 599 | { |
3d14c5d2 YS |
600 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
601 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
602 | struct ceph_mds_request *req; |
603 | int op; | |
604 | int err; | |
605 | ||
606 | dout("lookup %p dentry %p '%.*s'\n", | |
607 | dir, dentry, dentry->d_name.len, dentry->d_name.name); | |
608 | ||
609 | if (dentry->d_name.len > NAME_MAX) | |
610 | return ERR_PTR(-ENAMETOOLONG); | |
611 | ||
612 | err = ceph_init_dentry(dentry); | |
613 | if (err < 0) | |
614 | return ERR_PTR(err); | |
615 | ||
2817b000 SW |
616 | /* can we conclude ENOENT locally? */ |
617 | if (dentry->d_inode == NULL) { | |
618 | struct ceph_inode_info *ci = ceph_inode(dir); | |
619 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
620 | ||
be655596 | 621 | spin_lock(&ci->i_ceph_lock); |
2817b000 SW |
622 | dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags); |
623 | if (strncmp(dentry->d_name.name, | |
3d14c5d2 | 624 | fsc->mount_options->snapdir_name, |
2817b000 | 625 | dentry->d_name.len) && |
1d1de916 | 626 | !is_root_ceph_dentry(dir, dentry) && |
2f276c51 | 627 | __ceph_dir_is_complete(ci) && |
2817b000 | 628 | (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) { |
be655596 | 629 | spin_unlock(&ci->i_ceph_lock); |
2817b000 SW |
630 | dout(" dir %p complete, -ENOENT\n", dir); |
631 | d_add(dentry, NULL); | |
632 | di->lease_shared_gen = ci->i_shared_gen; | |
633 | return NULL; | |
634 | } | |
be655596 | 635 | spin_unlock(&ci->i_ceph_lock); |
2817b000 SW |
636 | } |
637 | ||
638 | op = ceph_snap(dir) == CEPH_SNAPDIR ? | |
639 | CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; | |
640 | req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS); | |
641 | if (IS_ERR(req)) | |
7e34bc52 | 642 | return ERR_CAST(req); |
2817b000 SW |
643 | req->r_dentry = dget(dentry); |
644 | req->r_num_caps = 2; | |
645 | /* we only need inode linkage */ | |
646 | req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE); | |
647 | req->r_locked_dir = dir; | |
648 | err = ceph_mdsc_do_request(mdsc, NULL, req); | |
468640e3 | 649 | err = ceph_handle_snapdir(req, dentry, err); |
2817b000 SW |
650 | dentry = ceph_finish_lookup(req, dentry, err); |
651 | ceph_mdsc_put_request(req); /* will dput(dentry) */ | |
652 | dout("lookup result=%p\n", dentry); | |
653 | return dentry; | |
654 | } | |
655 | ||
656 | /* | |
657 | * If we do a create but get no trace back from the MDS, follow up with | |
658 | * a lookup (the VFS expects us to link up the provided dentry). | |
659 | */ | |
660 | int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry) | |
661 | { | |
00cd8dd3 | 662 | struct dentry *result = ceph_lookup(dir, dentry, 0); |
2817b000 SW |
663 | |
664 | if (result && !IS_ERR(result)) { | |
665 | /* | |
666 | * We created the item, then did a lookup, and found | |
667 | * it was already linked to another inode we already | |
668 | * had in our cache (and thus got spliced). Link our | |
669 | * dentry to that inode, but don't hash it, just in | |
670 | * case the VFS wants to dereference it. | |
671 | */ | |
672 | BUG_ON(!result->d_inode); | |
673 | d_instantiate(dentry, result->d_inode); | |
674 | return 0; | |
675 | } | |
676 | return PTR_ERR(result); | |
677 | } | |
678 | ||
679 | static int ceph_mknod(struct inode *dir, struct dentry *dentry, | |
1a67aafb | 680 | umode_t mode, dev_t rdev) |
2817b000 | 681 | { |
3d14c5d2 YS |
682 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
683 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 | 684 | struct ceph_mds_request *req; |
b1ee94aa | 685 | struct ceph_acls_info acls = {}; |
2817b000 SW |
686 | int err; |
687 | ||
688 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
689 | return -EROFS; | |
690 | ||
b1ee94aa YZ |
691 | err = ceph_pre_init_acls(dir, &mode, &acls); |
692 | if (err < 0) | |
693 | return err; | |
694 | ||
1a67aafb | 695 | dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n", |
2817b000 SW |
696 | dir, dentry, mode, rdev); |
697 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS); | |
698 | if (IS_ERR(req)) { | |
b1ee94aa YZ |
699 | err = PTR_ERR(req); |
700 | goto out; | |
2817b000 SW |
701 | } |
702 | req->r_dentry = dget(dentry); | |
703 | req->r_num_caps = 2; | |
704 | req->r_locked_dir = dir; | |
705 | req->r_args.mknod.mode = cpu_to_le32(mode); | |
706 | req->r_args.mknod.rdev = cpu_to_le32(rdev); | |
707 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
708 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
b1ee94aa YZ |
709 | if (acls.pagelist) { |
710 | req->r_pagelist = acls.pagelist; | |
711 | acls.pagelist = NULL; | |
712 | } | |
2817b000 SW |
713 | err = ceph_mdsc_do_request(mdsc, dir, req); |
714 | if (!err && !req->r_reply_info.head->is_dentry) | |
715 | err = ceph_handle_notrace_create(dir, dentry); | |
716 | ceph_mdsc_put_request(req); | |
b1ee94aa | 717 | out: |
7221fe4c | 718 | if (!err) |
b1ee94aa | 719 | ceph_init_inode_acls(dentry->d_inode, &acls); |
b20a95a0 | 720 | else |
2817b000 | 721 | d_drop(dentry); |
b1ee94aa | 722 | ceph_release_acls_info(&acls); |
2817b000 SW |
723 | return err; |
724 | } | |
725 | ||
4acdaf27 | 726 | static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 727 | bool excl) |
2817b000 | 728 | { |
2d83bde9 | 729 | return ceph_mknod(dir, dentry, mode, 0); |
2817b000 SW |
730 | } |
731 | ||
732 | static int ceph_symlink(struct inode *dir, struct dentry *dentry, | |
733 | const char *dest) | |
734 | { | |
3d14c5d2 YS |
735 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
736 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
737 | struct ceph_mds_request *req; |
738 | int err; | |
739 | ||
740 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
741 | return -EROFS; | |
742 | ||
743 | dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest); | |
744 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS); | |
745 | if (IS_ERR(req)) { | |
b1ee94aa YZ |
746 | err = PTR_ERR(req); |
747 | goto out; | |
2817b000 SW |
748 | } |
749 | req->r_dentry = dget(dentry); | |
750 | req->r_num_caps = 2; | |
751 | req->r_path2 = kstrdup(dest, GFP_NOFS); | |
752 | req->r_locked_dir = dir; | |
753 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
754 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
755 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
756 | if (!err && !req->r_reply_info.head->is_dentry) | |
757 | err = ceph_handle_notrace_create(dir, dentry); | |
758 | ceph_mdsc_put_request(req); | |
b1ee94aa YZ |
759 | out: |
760 | if (err) | |
2817b000 SW |
761 | d_drop(dentry); |
762 | return err; | |
763 | } | |
764 | ||
18bb1db3 | 765 | static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
2817b000 | 766 | { |
3d14c5d2 YS |
767 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
768 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 | 769 | struct ceph_mds_request *req; |
b1ee94aa | 770 | struct ceph_acls_info acls = {}; |
2817b000 SW |
771 | int err = -EROFS; |
772 | int op; | |
773 | ||
774 | if (ceph_snap(dir) == CEPH_SNAPDIR) { | |
775 | /* mkdir .snap/foo is a MKSNAP */ | |
776 | op = CEPH_MDS_OP_MKSNAP; | |
777 | dout("mksnap dir %p snap '%.*s' dn %p\n", dir, | |
778 | dentry->d_name.len, dentry->d_name.name, dentry); | |
779 | } else if (ceph_snap(dir) == CEPH_NOSNAP) { | |
18bb1db3 | 780 | dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode); |
2817b000 SW |
781 | op = CEPH_MDS_OP_MKDIR; |
782 | } else { | |
783 | goto out; | |
784 | } | |
b1ee94aa YZ |
785 | |
786 | mode |= S_IFDIR; | |
787 | err = ceph_pre_init_acls(dir, &mode, &acls); | |
788 | if (err < 0) | |
789 | goto out; | |
790 | ||
2817b000 SW |
791 | req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); |
792 | if (IS_ERR(req)) { | |
793 | err = PTR_ERR(req); | |
794 | goto out; | |
795 | } | |
796 | ||
797 | req->r_dentry = dget(dentry); | |
798 | req->r_num_caps = 2; | |
799 | req->r_locked_dir = dir; | |
800 | req->r_args.mkdir.mode = cpu_to_le32(mode); | |
801 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
802 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
b1ee94aa YZ |
803 | if (acls.pagelist) { |
804 | req->r_pagelist = acls.pagelist; | |
805 | acls.pagelist = NULL; | |
806 | } | |
2817b000 SW |
807 | err = ceph_mdsc_do_request(mdsc, dir, req); |
808 | if (!err && !req->r_reply_info.head->is_dentry) | |
809 | err = ceph_handle_notrace_create(dir, dentry); | |
810 | ceph_mdsc_put_request(req); | |
811 | out: | |
b20a95a0 | 812 | if (!err) |
b1ee94aa | 813 | ceph_init_inode_acls(dentry->d_inode, &acls); |
b20a95a0 | 814 | else |
2817b000 | 815 | d_drop(dentry); |
b1ee94aa | 816 | ceph_release_acls_info(&acls); |
2817b000 SW |
817 | return err; |
818 | } | |
819 | ||
820 | static int ceph_link(struct dentry *old_dentry, struct inode *dir, | |
821 | struct dentry *dentry) | |
822 | { | |
3d14c5d2 YS |
823 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
824 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
825 | struct ceph_mds_request *req; |
826 | int err; | |
827 | ||
828 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
829 | return -EROFS; | |
830 | ||
831 | dout("link in dir %p old_dentry %p dentry %p\n", dir, | |
832 | old_dentry, dentry); | |
833 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS); | |
834 | if (IS_ERR(req)) { | |
835 | d_drop(dentry); | |
836 | return PTR_ERR(req); | |
837 | } | |
838 | req->r_dentry = dget(dentry); | |
839 | req->r_num_caps = 2; | |
4b58c9b1 | 840 | req->r_old_dentry = dget(old_dentry); |
2817b000 SW |
841 | req->r_locked_dir = dir; |
842 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
843 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
ad88f23f YZ |
844 | /* release LINK_SHARED on source inode (mds will lock it) */ |
845 | req->r_old_inode_drop = CEPH_CAP_LINK_SHARED; | |
2817b000 | 846 | err = ceph_mdsc_do_request(mdsc, dir, req); |
70b666c3 | 847 | if (err) { |
2817b000 | 848 | d_drop(dentry); |
70b666c3 SW |
849 | } else if (!req->r_reply_info.head->is_dentry) { |
850 | ihold(old_dentry->d_inode); | |
851 | d_instantiate(dentry, old_dentry->d_inode); | |
852 | } | |
2817b000 SW |
853 | ceph_mdsc_put_request(req); |
854 | return err; | |
855 | } | |
856 | ||
857 | /* | |
858 | * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it | |
859 | * looks like the link count will hit 0, drop any other caps (other | |
860 | * than PIN) we don't specifically want (due to the file still being | |
861 | * open). | |
862 | */ | |
863 | static int drop_caps_for_unlink(struct inode *inode) | |
864 | { | |
865 | struct ceph_inode_info *ci = ceph_inode(inode); | |
866 | int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; | |
867 | ||
be655596 | 868 | spin_lock(&ci->i_ceph_lock); |
2817b000 SW |
869 | if (inode->i_nlink == 1) { |
870 | drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN); | |
871 | ci->i_ceph_flags |= CEPH_I_NODELAY; | |
872 | } | |
be655596 | 873 | spin_unlock(&ci->i_ceph_lock); |
2817b000 SW |
874 | return drop; |
875 | } | |
876 | ||
877 | /* | |
878 | * rmdir and unlink are differ only by the metadata op code | |
879 | */ | |
880 | static int ceph_unlink(struct inode *dir, struct dentry *dentry) | |
881 | { | |
3d14c5d2 YS |
882 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
883 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
884 | struct inode *inode = dentry->d_inode; |
885 | struct ceph_mds_request *req; | |
886 | int err = -EROFS; | |
887 | int op; | |
888 | ||
889 | if (ceph_snap(dir) == CEPH_SNAPDIR) { | |
890 | /* rmdir .snap/foo is RMSNAP */ | |
891 | dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len, | |
892 | dentry->d_name.name, dentry); | |
893 | op = CEPH_MDS_OP_RMSNAP; | |
894 | } else if (ceph_snap(dir) == CEPH_NOSNAP) { | |
895 | dout("unlink/rmdir dir %p dn %p inode %p\n", | |
896 | dir, dentry, inode); | |
dba19c60 | 897 | op = S_ISDIR(dentry->d_inode->i_mode) ? |
2817b000 SW |
898 | CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK; |
899 | } else | |
900 | goto out; | |
901 | req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); | |
902 | if (IS_ERR(req)) { | |
903 | err = PTR_ERR(req); | |
904 | goto out; | |
905 | } | |
906 | req->r_dentry = dget(dentry); | |
907 | req->r_num_caps = 2; | |
908 | req->r_locked_dir = dir; | |
909 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
910 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
911 | req->r_inode_drop = drop_caps_for_unlink(inode); | |
912 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
913 | if (!err && !req->r_reply_info.head->is_dentry) | |
914 | d_delete(dentry); | |
915 | ceph_mdsc_put_request(req); | |
916 | out: | |
917 | return err; | |
918 | } | |
919 | ||
920 | static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry, | |
921 | struct inode *new_dir, struct dentry *new_dentry) | |
922 | { | |
3d14c5d2 YS |
923 | struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb); |
924 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
925 | struct ceph_mds_request *req; |
926 | int err; | |
927 | ||
928 | if (ceph_snap(old_dir) != ceph_snap(new_dir)) | |
929 | return -EXDEV; | |
930 | if (ceph_snap(old_dir) != CEPH_NOSNAP || | |
931 | ceph_snap(new_dir) != CEPH_NOSNAP) | |
932 | return -EROFS; | |
933 | dout("rename dir %p dentry %p to dir %p dentry %p\n", | |
934 | old_dir, old_dentry, new_dir, new_dentry); | |
935 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS); | |
936 | if (IS_ERR(req)) | |
937 | return PTR_ERR(req); | |
180061a5 | 938 | ihold(old_dir); |
2817b000 SW |
939 | req->r_dentry = dget(new_dentry); |
940 | req->r_num_caps = 2; | |
941 | req->r_old_dentry = dget(old_dentry); | |
180061a5 | 942 | req->r_old_dentry_dir = old_dir; |
2817b000 SW |
943 | req->r_locked_dir = new_dir; |
944 | req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED; | |
945 | req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL; | |
946 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
947 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
948 | /* release LINK_RDCACHE on source inode (mds will lock it) */ | |
949 | req->r_old_inode_drop = CEPH_CAP_LINK_SHARED; | |
950 | if (new_dentry->d_inode) | |
951 | req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode); | |
952 | err = ceph_mdsc_do_request(mdsc, old_dir, req); | |
953 | if (!err && !req->r_reply_info.head->is_dentry) { | |
954 | /* | |
955 | * Normally d_move() is done by fill_trace (called by | |
956 | * do_request, above). If there is no trace, we need | |
957 | * to do it here. | |
958 | */ | |
ea1409f9 | 959 | |
2817b000 | 960 | d_move(old_dentry, new_dentry); |
ea1409f9 SW |
961 | |
962 | /* ensure target dentry is invalidated, despite | |
963 | rehashing bug in vfs_rename_dir */ | |
81a6cf2d | 964 | ceph_invalidate_dentry_lease(new_dentry); |
0a8a70f9 YZ |
965 | |
966 | /* d_move screws up sibling dentries' offsets */ | |
967 | ceph_dir_clear_complete(old_dir); | |
968 | ceph_dir_clear_complete(new_dir); | |
969 | ||
2817b000 SW |
970 | } |
971 | ceph_mdsc_put_request(req); | |
972 | return err; | |
973 | } | |
974 | ||
81a6cf2d SW |
975 | /* |
976 | * Ensure a dentry lease will no longer revalidate. | |
977 | */ | |
978 | void ceph_invalidate_dentry_lease(struct dentry *dentry) | |
979 | { | |
980 | spin_lock(&dentry->d_lock); | |
981 | dentry->d_time = jiffies; | |
982 | ceph_dentry(dentry)->lease_shared_gen = 0; | |
983 | spin_unlock(&dentry->d_lock); | |
984 | } | |
2817b000 SW |
985 | |
986 | /* | |
987 | * Check if dentry lease is valid. If not, delete the lease. Try to | |
988 | * renew if the least is more than half up. | |
989 | */ | |
990 | static int dentry_lease_is_valid(struct dentry *dentry) | |
991 | { | |
992 | struct ceph_dentry_info *di; | |
993 | struct ceph_mds_session *s; | |
994 | int valid = 0; | |
995 | u32 gen; | |
996 | unsigned long ttl; | |
997 | struct ceph_mds_session *session = NULL; | |
998 | struct inode *dir = NULL; | |
999 | u32 seq = 0; | |
1000 | ||
1001 | spin_lock(&dentry->d_lock); | |
1002 | di = ceph_dentry(dentry); | |
3d8eb7a9 | 1003 | if (di->lease_session) { |
2817b000 | 1004 | s = di->lease_session; |
d8fb02ab | 1005 | spin_lock(&s->s_gen_ttl_lock); |
2817b000 SW |
1006 | gen = s->s_cap_gen; |
1007 | ttl = s->s_cap_ttl; | |
d8fb02ab | 1008 | spin_unlock(&s->s_gen_ttl_lock); |
2817b000 SW |
1009 | |
1010 | if (di->lease_gen == gen && | |
1011 | time_before(jiffies, dentry->d_time) && | |
1012 | time_before(jiffies, ttl)) { | |
1013 | valid = 1; | |
1014 | if (di->lease_renew_after && | |
1015 | time_after(jiffies, di->lease_renew_after)) { | |
1016 | /* we should renew */ | |
1017 | dir = dentry->d_parent->d_inode; | |
1018 | session = ceph_get_mds_session(s); | |
1019 | seq = di->lease_seq; | |
1020 | di->lease_renew_after = 0; | |
1021 | di->lease_renew_from = jiffies; | |
1022 | } | |
2817b000 SW |
1023 | } |
1024 | } | |
1025 | spin_unlock(&dentry->d_lock); | |
1026 | ||
1027 | if (session) { | |
1028 | ceph_mdsc_lease_send_msg(session, dir, dentry, | |
1029 | CEPH_MDS_LEASE_RENEW, seq); | |
1030 | ceph_put_mds_session(session); | |
1031 | } | |
1032 | dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid); | |
1033 | return valid; | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * Check if directory-wide content lease/cap is valid. | |
1038 | */ | |
1039 | static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry) | |
1040 | { | |
1041 | struct ceph_inode_info *ci = ceph_inode(dir); | |
1042 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
1043 | int valid = 0; | |
1044 | ||
be655596 | 1045 | spin_lock(&ci->i_ceph_lock); |
2817b000 SW |
1046 | if (ci->i_shared_gen == di->lease_shared_gen) |
1047 | valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1); | |
be655596 | 1048 | spin_unlock(&ci->i_ceph_lock); |
2817b000 SW |
1049 | dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n", |
1050 | dir, (unsigned)ci->i_shared_gen, dentry, | |
1051 | (unsigned)di->lease_shared_gen, valid); | |
1052 | return valid; | |
1053 | } | |
1054 | ||
1055 | /* | |
1056 | * Check if cached dentry can be trusted. | |
1057 | */ | |
0b728e19 | 1058 | static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags) |
2817b000 | 1059 | { |
bf1c6aca | 1060 | int valid = 0; |
34286d66 NP |
1061 | struct inode *dir; |
1062 | ||
0b728e19 | 1063 | if (flags & LOOKUP_RCU) |
34286d66 NP |
1064 | return -ECHILD; |
1065 | ||
1cd3935b SW |
1066 | dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry, |
1067 | dentry->d_name.len, dentry->d_name.name, dentry->d_inode, | |
1068 | ceph_dentry(dentry)->offset); | |
2817b000 | 1069 | |
bf1c6aca SW |
1070 | dir = ceph_get_dentry_parent_inode(dentry); |
1071 | ||
2817b000 SW |
1072 | /* always trust cached snapped dentries, snapdir dentry */ |
1073 | if (ceph_snap(dir) != CEPH_NOSNAP) { | |
1074 | dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry, | |
1075 | dentry->d_name.len, dentry->d_name.name, dentry->d_inode); | |
bf1c6aca SW |
1076 | valid = 1; |
1077 | } else if (dentry->d_inode && | |
1078 | ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) { | |
1079 | valid = 1; | |
1080 | } else if (dentry_lease_is_valid(dentry) || | |
1081 | dir_lease_is_valid(dir, dentry)) { | |
9215aeea YZ |
1082 | if (dentry->d_inode) |
1083 | valid = ceph_is_any_caps(dentry->d_inode); | |
1084 | else | |
1085 | valid = 1; | |
2817b000 | 1086 | } |
2817b000 | 1087 | |
bf1c6aca | 1088 | dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid"); |
9215aeea | 1089 | if (valid) { |
bf1c6aca | 1090 | ceph_dentry_lru_touch(dentry); |
9215aeea YZ |
1091 | } else { |
1092 | ceph_dir_clear_complete(dir); | |
9215aeea | 1093 | } |
bf1c6aca SW |
1094 | iput(dir); |
1095 | return valid; | |
2817b000 SW |
1096 | } |
1097 | ||
1098 | /* | |
147851d2 | 1099 | * Release our ceph_dentry_info. |
2817b000 | 1100 | */ |
147851d2 | 1101 | static void ceph_d_release(struct dentry *dentry) |
2817b000 SW |
1102 | { |
1103 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
2817b000 | 1104 | |
147851d2 | 1105 | dout("d_release %p\n", dentry); |
3d8eb7a9 SW |
1106 | ceph_dentry_lru_del(dentry); |
1107 | if (di->lease_session) | |
1108 | ceph_put_mds_session(di->lease_session); | |
1109 | kmem_cache_free(ceph_dentry_cachep, di); | |
1110 | dentry->d_fsdata = NULL; | |
2817b000 SW |
1111 | } |
1112 | ||
1113 | static int ceph_snapdir_d_revalidate(struct dentry *dentry, | |
0b728e19 | 1114 | unsigned int flags) |
2817b000 SW |
1115 | { |
1116 | /* | |
1117 | * Eventually, we'll want to revalidate snapped metadata | |
1118 | * too... probably... | |
1119 | */ | |
1120 | return 1; | |
1121 | } | |
1122 | ||
b58dc410 SW |
1123 | /* |
1124 | * When the VFS prunes a dentry from the cache, we need to clear the | |
1125 | * complete flag on the parent directory. | |
1126 | * | |
1127 | * Called under dentry->d_lock. | |
1128 | */ | |
1129 | static void ceph_d_prune(struct dentry *dentry) | |
1130 | { | |
774ac21d | 1131 | dout("ceph_d_prune %p\n", dentry); |
b58dc410 SW |
1132 | |
1133 | /* do we have a valid parent? */ | |
8842b3be | 1134 | if (IS_ROOT(dentry)) |
b58dc410 SW |
1135 | return; |
1136 | ||
2f276c51 | 1137 | /* if we are not hashed, we don't affect dir's completeness */ |
b58dc410 SW |
1138 | if (d_unhashed(dentry)) |
1139 | return; | |
2817b000 | 1140 | |
b58dc410 SW |
1141 | /* |
1142 | * we hold d_lock, so d_parent is stable, and d_fsdata is never | |
1143 | * cleared until d_release | |
1144 | */ | |
2f276c51 | 1145 | ceph_dir_clear_complete(dentry->d_parent->d_inode); |
b58dc410 | 1146 | } |
2817b000 SW |
1147 | |
1148 | /* | |
1149 | * read() on a dir. This weird interface hack only works if mounted | |
1150 | * with '-o dirstat'. | |
1151 | */ | |
1152 | static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size, | |
1153 | loff_t *ppos) | |
1154 | { | |
1155 | struct ceph_file_info *cf = file->private_data; | |
496ad9aa | 1156 | struct inode *inode = file_inode(file); |
2817b000 SW |
1157 | struct ceph_inode_info *ci = ceph_inode(inode); |
1158 | int left; | |
ae598083 | 1159 | const int bufsize = 1024; |
2817b000 | 1160 | |
3d14c5d2 | 1161 | if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT)) |
2817b000 SW |
1162 | return -EISDIR; |
1163 | ||
1164 | if (!cf->dir_info) { | |
ae598083 | 1165 | cf->dir_info = kmalloc(bufsize, GFP_NOFS); |
2817b000 SW |
1166 | if (!cf->dir_info) |
1167 | return -ENOMEM; | |
1168 | cf->dir_info_len = | |
ae598083 | 1169 | snprintf(cf->dir_info, bufsize, |
2817b000 SW |
1170 | "entries: %20lld\n" |
1171 | " files: %20lld\n" | |
1172 | " subdirs: %20lld\n" | |
1173 | "rentries: %20lld\n" | |
1174 | " rfiles: %20lld\n" | |
1175 | " rsubdirs: %20lld\n" | |
1176 | "rbytes: %20lld\n" | |
1177 | "rctime: %10ld.%09ld\n", | |
1178 | ci->i_files + ci->i_subdirs, | |
1179 | ci->i_files, | |
1180 | ci->i_subdirs, | |
1181 | ci->i_rfiles + ci->i_rsubdirs, | |
1182 | ci->i_rfiles, | |
1183 | ci->i_rsubdirs, | |
1184 | ci->i_rbytes, | |
1185 | (long)ci->i_rctime.tv_sec, | |
1186 | (long)ci->i_rctime.tv_nsec); | |
1187 | } | |
1188 | ||
1189 | if (*ppos >= cf->dir_info_len) | |
1190 | return 0; | |
1191 | size = min_t(unsigned, size, cf->dir_info_len-*ppos); | |
1192 | left = copy_to_user(buf, cf->dir_info + *ppos, size); | |
1193 | if (left == size) | |
1194 | return -EFAULT; | |
1195 | *ppos += (size - left); | |
1196 | return size - left; | |
1197 | } | |
1198 | ||
1199 | /* | |
1200 | * an fsync() on a dir will wait for any uncommitted directory | |
1201 | * operations to commit. | |
1202 | */ | |
02c24a82 JB |
1203 | static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end, |
1204 | int datasync) | |
2817b000 | 1205 | { |
496ad9aa | 1206 | struct inode *inode = file_inode(file); |
2817b000 SW |
1207 | struct ceph_inode_info *ci = ceph_inode(inode); |
1208 | struct list_head *head = &ci->i_unsafe_dirops; | |
1209 | struct ceph_mds_request *req; | |
1210 | u64 last_tid; | |
1211 | int ret = 0; | |
1212 | ||
1213 | dout("dir_fsync %p\n", inode); | |
02c24a82 JB |
1214 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); |
1215 | if (ret) | |
1216 | return ret; | |
1217 | mutex_lock(&inode->i_mutex); | |
1218 | ||
2817b000 SW |
1219 | spin_lock(&ci->i_unsafe_lock); |
1220 | if (list_empty(head)) | |
1221 | goto out; | |
1222 | ||
1223 | req = list_entry(head->prev, | |
1224 | struct ceph_mds_request, r_unsafe_dir_item); | |
1225 | last_tid = req->r_tid; | |
1226 | ||
1227 | do { | |
1228 | ceph_mdsc_get_request(req); | |
1229 | spin_unlock(&ci->i_unsafe_lock); | |
2ff179e6 | 1230 | |
2817b000 SW |
1231 | dout("dir_fsync %p wait on tid %llu (until %llu)\n", |
1232 | inode, req->r_tid, last_tid); | |
1233 | if (req->r_timeout) { | |
1234 | ret = wait_for_completion_timeout( | |
1235 | &req->r_safe_completion, req->r_timeout); | |
1236 | if (ret > 0) | |
1237 | ret = 0; | |
1238 | else if (ret == 0) | |
1239 | ret = -EIO; /* timed out */ | |
1240 | } else { | |
1241 | wait_for_completion(&req->r_safe_completion); | |
1242 | } | |
2817b000 SW |
1243 | ceph_mdsc_put_request(req); |
1244 | ||
2ff179e6 | 1245 | spin_lock(&ci->i_unsafe_lock); |
2817b000 SW |
1246 | if (ret || list_empty(head)) |
1247 | break; | |
1248 | req = list_entry(head->next, | |
1249 | struct ceph_mds_request, r_unsafe_dir_item); | |
1250 | } while (req->r_tid < last_tid); | |
1251 | out: | |
1252 | spin_unlock(&ci->i_unsafe_lock); | |
02c24a82 JB |
1253 | mutex_unlock(&inode->i_mutex); |
1254 | ||
2817b000 SW |
1255 | return ret; |
1256 | } | |
1257 | ||
1258 | /* | |
1259 | * We maintain a private dentry LRU. | |
1260 | * | |
1261 | * FIXME: this needs to be changed to a per-mds lru to be useful. | |
1262 | */ | |
1263 | void ceph_dentry_lru_add(struct dentry *dn) | |
1264 | { | |
1265 | struct ceph_dentry_info *di = ceph_dentry(dn); | |
1266 | struct ceph_mds_client *mdsc; | |
2817b000 | 1267 | |
04a419f9 SW |
1268 | dout("dentry_lru_add %p %p '%.*s'\n", di, dn, |
1269 | dn->d_name.len, dn->d_name.name); | |
3d8eb7a9 SW |
1270 | mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; |
1271 | spin_lock(&mdsc->dentry_lru_lock); | |
1272 | list_add_tail(&di->lru, &mdsc->dentry_lru); | |
1273 | mdsc->num_dentry++; | |
1274 | spin_unlock(&mdsc->dentry_lru_lock); | |
2817b000 SW |
1275 | } |
1276 | ||
1277 | void ceph_dentry_lru_touch(struct dentry *dn) | |
1278 | { | |
1279 | struct ceph_dentry_info *di = ceph_dentry(dn); | |
1280 | struct ceph_mds_client *mdsc; | |
2817b000 | 1281 | |
1cd3935b SW |
1282 | dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn, |
1283 | dn->d_name.len, dn->d_name.name, di->offset); | |
3d8eb7a9 SW |
1284 | mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; |
1285 | spin_lock(&mdsc->dentry_lru_lock); | |
1286 | list_move_tail(&di->lru, &mdsc->dentry_lru); | |
1287 | spin_unlock(&mdsc->dentry_lru_lock); | |
2817b000 SW |
1288 | } |
1289 | ||
1290 | void ceph_dentry_lru_del(struct dentry *dn) | |
1291 | { | |
1292 | struct ceph_dentry_info *di = ceph_dentry(dn); | |
1293 | struct ceph_mds_client *mdsc; | |
1294 | ||
04a419f9 SW |
1295 | dout("dentry_lru_del %p %p '%.*s'\n", di, dn, |
1296 | dn->d_name.len, dn->d_name.name); | |
3d8eb7a9 SW |
1297 | mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; |
1298 | spin_lock(&mdsc->dentry_lru_lock); | |
1299 | list_del_init(&di->lru); | |
1300 | mdsc->num_dentry--; | |
1301 | spin_unlock(&mdsc->dentry_lru_lock); | |
2817b000 SW |
1302 | } |
1303 | ||
6c0f3af7 SW |
1304 | /* |
1305 | * Return name hash for a given dentry. This is dependent on | |
1306 | * the parent directory's hash function. | |
1307 | */ | |
e5f86dc3 | 1308 | unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn) |
6c0f3af7 | 1309 | { |
6c0f3af7 SW |
1310 | struct ceph_inode_info *dci = ceph_inode(dir); |
1311 | ||
1312 | switch (dci->i_dir_layout.dl_dir_hash) { | |
1313 | case 0: /* for backward compat */ | |
1314 | case CEPH_STR_HASH_LINUX: | |
1315 | return dn->d_name.hash; | |
1316 | ||
1317 | default: | |
1318 | return ceph_str_hash(dci->i_dir_layout.dl_dir_hash, | |
1319 | dn->d_name.name, dn->d_name.len); | |
1320 | } | |
1321 | } | |
1322 | ||
2817b000 SW |
1323 | const struct file_operations ceph_dir_fops = { |
1324 | .read = ceph_read_dir, | |
77acfa29 | 1325 | .iterate = ceph_readdir, |
2817b000 SW |
1326 | .llseek = ceph_dir_llseek, |
1327 | .open = ceph_open, | |
1328 | .release = ceph_release, | |
1329 | .unlocked_ioctl = ceph_ioctl, | |
1330 | .fsync = ceph_dir_fsync, | |
1331 | }; | |
1332 | ||
1333 | const struct inode_operations ceph_dir_iops = { | |
1334 | .lookup = ceph_lookup, | |
1335 | .permission = ceph_permission, | |
1336 | .getattr = ceph_getattr, | |
1337 | .setattr = ceph_setattr, | |
1338 | .setxattr = ceph_setxattr, | |
1339 | .getxattr = ceph_getxattr, | |
1340 | .listxattr = ceph_listxattr, | |
1341 | .removexattr = ceph_removexattr, | |
7221fe4c | 1342 | .get_acl = ceph_get_acl, |
72466d0b | 1343 | .set_acl = ceph_set_acl, |
2817b000 SW |
1344 | .mknod = ceph_mknod, |
1345 | .symlink = ceph_symlink, | |
1346 | .mkdir = ceph_mkdir, | |
1347 | .link = ceph_link, | |
1348 | .unlink = ceph_unlink, | |
1349 | .rmdir = ceph_unlink, | |
1350 | .rename = ceph_rename, | |
1351 | .create = ceph_create, | |
2d83bde9 | 1352 | .atomic_open = ceph_atomic_open, |
2817b000 SW |
1353 | }; |
1354 | ||
52dfb8ac | 1355 | const struct dentry_operations ceph_dentry_ops = { |
2817b000 | 1356 | .d_revalidate = ceph_d_revalidate, |
147851d2 | 1357 | .d_release = ceph_d_release, |
b58dc410 | 1358 | .d_prune = ceph_d_prune, |
2817b000 SW |
1359 | }; |
1360 | ||
52dfb8ac | 1361 | const struct dentry_operations ceph_snapdir_dentry_ops = { |
2817b000 | 1362 | .d_revalidate = ceph_snapdir_d_revalidate, |
147851d2 | 1363 | .d_release = ceph_d_release, |
2817b000 SW |
1364 | }; |
1365 | ||
52dfb8ac | 1366 | const struct dentry_operations ceph_snap_dentry_ops = { |
147851d2 | 1367 | .d_release = ceph_d_release, |
b58dc410 | 1368 | .d_prune = ceph_d_prune, |
2817b000 | 1369 | }; |